Browse Prior Art Database

Lossless/ Lossy Transform Image Coding with Extended Context Algorithm

IP.com Disclosure Number: IPCOM000110676D
Original Publication Date: 1992-Dec-01
Included in the Prior Art Database: 2005-Mar-25
Document File: 2 page(s) / 82K

Publishing Venue

IBM

Related People

Furlan, G: AUTHOR [+2]

Abstract

This article describes an improvement of Algorithm Context [1,2,3] which enhances its model building performance whenever the observed data range over a large alphabet. Of particular interest is the application to lossy and lossless data compression of images based upon either the technique of the so-called subband coding [4] or the use of Discrete Cosine Transform (DCT) [5]. Both of these techniques provide a potentially powerful way to achieve compression, but at the same time there is an increase of the required alphabet as well as another adverse data reduction phenomenon which together make it very difficult to take full advantage of their inherent power. The disclosed version of the algorithm Context overcomes both of the difficulties.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Lossless/ Lossy Transform Image Coding with Extended Context Algorithm

       This article describes an improvement of Algorithm
Context [1,2,3] which enhances its model building performance
whenever the observed data range over a large alphabet.  Of
particular interest is the application to lossy and lossless data
compression of images based upon either the technique of the
so-called subband coding [4] or the use of Discrete Cosine Transform
(DCT) [5].  Both of these techniques provide a potentially powerful
way to achieve compression, but at the same time there is an increase
of the required alphabet as well as another adverse data reduction
phenomenon which together make it very difficult to take full
advantage of their inherent power.  The disclosed version of the
algorithm Context overcomes both of the difficulties.

      The code length generated by the original version of Algorithm
Context can be expressed as follows:

                            (Image Omitted)

where f is a reordering function of the past symbols which are taken
in a decreasing order of influence.  A specific function f is
selected for each subband image.  This creates a context by
reordering the past pixels of the current subband image as well as of
the other subbands.

      For the purposes of context creation this function is now
applied to suitably quantized values of the past pixels, and the code
length can then be expressed as

                            (Image Omitted)

where xq denotes the number of distinct quantized symbols permitted.
This still leaves the freedom of either quantizing the next xt for
lossy compression or leaving it to the full precision to achieve
lossless compression.  In either case, a grouping of contexts is
achieved which are close to each other to reduce their total number
in...